- Email: [email protected]
Validating a centralized approach to primary frequency control with series-produced electric vehicles
Journal of Energy Storage 7 (2016) 63–73
Contents lists available at ScienceDirect
Journal of Energy Storage journal homepage: www.elsevier.com/locate/est
Validating a centralized approach to primary frequency control with series-produced electric vehicles Mattia Marinelli* , Sergejus Martinenas, Katarina Knezovi c, Peter Bach Andersen Centre for Electric Power and Energy, Technical University of Denmark, Risø Campus, Roskilde, Denmark
A R T I C L E I N F O
Article history: Received 19 March 2016 Received in revised form 16 May 2016 Accepted 17 May 2016 Available online xxx Keywords: Centralized control Electric vehicles Frequency control Islanded systems Testing
A B S T R A C T
The aim of this work is twofold: on one hand it proposes a centralized approach to primary frequency control by using electric vehicles as controllable units; on the other hand, it experimentally validates whether series-produced EVs, adhering to contemporary standards, can be an effective resource for providing primary frequency control. The validation process is realized in an islanded system with renewable sources and it relies on verifying that the frequency values are within the desired limits following severe load steps or wind power fluctuations. In order to reflect today’s situation, the used EVs, three Nissan Leaf, are not taking advantage of any V2G capability, but rely solely on the possibility of limiting the charge between 6 A and 16 A. The centralized approach implies that the frequency is not measured locally as it is a common practice today, but is routed via the Internet in order to include potential communication delays that would take into account the presence of different entities for controlling the vehicles, such as aggregators and utilities. The centralised approach is pursued to support aggregators in participating in current ancillary service markets. Ultimately, this paper aims to strengthen the applied research within EV integration through the practical validation of smart grid concepts on original manufactured equipment. ã 2016 Elsevier Ltd. All rights reserved.
1. Introduction With conventional units being replaced by renewable resources, there is an increased demand for additional ancillary services, such as frequency control. Due to their defining property of being quick-response high-power units, electric vehicles (EVs) emerge as a viable actor for providing frequency regulation. A noticeable amount of research efforts is put nowadays to investigate all different aspects needed to shift from the traditional power system, where frequency and voltages are controlled by a relatively small set of large units, into a futuristic system, where potentially all power devices can be involved in the controlling
Abbreviations: DSO, distribution system operator; ENTSO-E, European Network of Transmission System Operators for Electricity; EV, electric vehicle; EVSE, electric vehicle supply equipment; FCR, frequency containment reserve; FRR, frequency restoration reserve; ICT, information and communications technology; OEM, original equipment manufacturer; OR, Operating reserve; PFC, primary frequency control; RES, renewable energy sources; RR, Replacement reserve; SOC, state-ofcharge; TSO, transmission system operator; V2G, vehicle to grid. * Corresponding author. E-mail addresses: [email protected], [email protected] (M. Marinelli). http://dx.doi.org/10.1016/j.est.2016.05.008 2352-152X/ã 2016 Elsevier Ltd. All rights reserved.
actions. For instance, in the ELECTRA project, innovative control schemes are being investigated in order to assess whether conventional frequency and voltage controlling approaches are still suitable, or which aspects need to be revised for scenarios with massive amounts of small distributed energy resources [1,2]. It is argued that TSOs alone will not be able to effectively manage the overall system balance and will have to partially delegate frequency control responsibility to DSOs as already happens today for congestions management and volt-reactive power provision [3–10]. Regardless of who will be responsible for frequency and voltage control, several technical challenges are ahead. For instance, will the aggregated provision of thousands of few kVA-units at low voltage level be as effective as the one of MVA-units response? Moreover, considering the traditional 3-phase system, will the response of groups of single-phase devices be as effective as the response of similar size 3-phase units? With a specific focus on frequency control, is it reasonable to expect millions of accurate local frequency measurements to be used for decentralized, traditional, droop controllers or is it rather better to have a limited set of centralized controllers, which rely on a few accurate
64
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
measurements, sending out power set-points via normal Internet connection? 1.1. Background and literature analysis Electric vehicles are one of the imminent candidates for providing ancillary services. Most of the time (typically about 90%) they are plugged into a charging post and can, in principle, provide fast-regulating power in both directions, or just modulate the charging power. A noticeable amount of literature has already been written supporting this statement [7–22]. It is argued that EVs with V2G capability can provide regulation services, and can compete in electricity markets, such as markets for ancillary services, where there is payment for available capacity, apart from the payment for the actual dispatch. Frequency control is one of the services which can be provided by EVs through this market. More specifically, primary frequency control (PFC) can be suitably provided by EVs due to their flexible operating mode and ability to seamlessly alter the consuming/ producing power under the V2G concept [12–14]. The work described in [15] presents an aggregated PFC model, where a participation factor, dependent on the state-of-charge (SOC), is associated with each EV to determine its droop characteristic. It was shown that EVs can effectively improve the system frequency response, as well as that V2G-capable vehicles have better power response, due to more available primary reserves. Furthermore, a decentralized V2G control for primary frequency regulation is presented in [16]. The proposed method considers customer charging demands and adapts the frequency droop control to maintain or achieve the desired SOC. A comparative study is performed in [17], in order to evaluate benefits of EVs performing primary frequency control in an islanded system with high penetration of renewable resources. The presented study case argues that system frequency oscillates in a 0.3 Hz band if the EVs contribute to primary regulation, compared to 1 Hz in the case of only using available hydro units. Two studies, respectively from Japan and Great Britain, analyse participation in frequency control on large systems considering traveling constraints and including large amount of renewable sources [18,19]. However, even though the mentioned studies analysed different strategies for providing primary frequency control, rarely have they dealt with the experimental validation, but mostly remained on modelling and simulations. For example, the works described in [15–21] have implemented different droop controls and shown that EVs can be effective in primary frequency regulation, likewise in isolated microgrids and larger systems. Still, they assume an ideal EV response to the control signals, both in the reaction time and the provided power, and they omit communication and control latencies which may greatly impact the results. In addition, technical challenges may arise due to the limited power and energy size of each individual unit, as well as the need to have simple and effective measuring and controlling capabilities for primary frequency regulation. Transmission systems operators may be sceptical about the possibility of having demand participating in the frequency regulation, mainly because of response uncertainties and metering inaccuracies. Therefore, an extensive experimental activity is required to prove the feasibility of these solutions. The activity described in this paper is carried out using series-produced vehicles and the universally supported IEC 61851 standard, to prove the applicability of the solution.
control, whereas the experimental validation is rarely touched upon. Therefore, this paper focuses on the evaluation of EVs’ ability to provide primary frequency control in a centralized fashion. The frequency control analysis of three EVs connected to a small islanded power system is proposed. Having the system islanded gives the possibility of emulating realistic frequency events, which could hardly be appreciated if connected to the national grid. It is important to note that the experiments are carried out with commercially available vehicles without taking advantage of any V2G capability, but only with the possibility to modulate the unidirectional charging current. This limit is part of internal standards (IEC61851/J1772) for conductive AC charging and supported by the vast majority of EVs today. A classical droop control function is utilized. However, contrary to today’s practice, which relies on local measurement, the frequency measurement is routed via the Internet to the controller, which sends the current set-point to the EV. The authors believe that in the near future, it will be highly unlikely to equip each EV with a precise measurement device which meets the TSO requirements. This means that the used technology resembles that of an operational environment: a fleet of EVs providing frequency regulation on market terms through an aggregator who has the certified frequency measurement device. Such setup allows the inclusion and assessment of potential communication delays, especially their influence on system stability. This work does not take in consideration vehicle unavailability due to owner usage. However it has to be reminded that, due to the system wide nature of the service, the location of the resource providing frequency control is not extremely important. In that sense, it is the aggregator’s best interest to rely on a higher number of vehicles to account for the plug-in uncertainties, whilst it is less important to know their exact location, as long as the utilized vehicles are connected to the grid and not used for other services. Ultimately, the research question tackled in this paper is: can small size, single phase distributed energy resources, such as commercially available electric vehicles, effectively provide primary frequency control relying on a centralized controller which sends out current set-points? The rest of the paper is structured as follows: Section 2 briefly recalls how the control of frequency is traditionally organized in Europe and in Denmark. Section 3 describes the controller characteristics, the communication architecture and the implementation in the laboratory. In Section 4 numerical and graphical results of the experiments are presented and discussed; several scenarios are investigated from load step, through steady state analysis to wind power balancing. Section 5 reports conclusions and lessons learned. 2. Current framework for frequency control in Europe and Denmark 2.1. Frequency services according to ENTSO-E division Based on the European Network of Transmission System Operators for Electricity (ENTSO-E) definitions, reported in the Network Code and Operation Handbook, frequency control includes [23]: Primary frequency control; Secondary power-frequency control; Tertiary control.
1.2. Objective of the manuscript Most of the literature identified during the literature review focuses on modeling and simulating the EV primary frequency
ENTSO-E refers to the reserves for frequency control as Operating Reserves (OR), and specifically, indicates the abovementioned controls as:
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
Frequency Containment Reserves (FCR); Frequency Restoration Reserves (FRR); Replacement Reserves (RR). The (automatic) primary frequency control aims at achieving the operational reliability of the synchronous area by stabilizing the system frequency after a disturbance or an incident at an acceptable stationary value in the second time frame, typically within 30 s. The minimum requirement for FCR is based on the maximum loss of power generation or demand, also called reference incident: within the Continental European synchronous area, the reference incident is defined as 3000 MW of generating capacity. FCR is shared among all control areas. The requirement of each control area is defined by contribution coefficients, calculated as the ratio between the energy generated in that area over a year and the energy generated in the entire synchronous area. The response has to be maintained for up to 15 min. The secondary frequency control, which can be either automatic or manual, aims to restore the system frequency within a few minutes, typically up to 15 min after incident, by releasing systemwide activated frequency containment reserves. For large interconnected systems, where a decentralized frequency restoration control is implemented, frequency restoration also aims at restoring the balance between generationand load foreach TSO, andconsequently, restores power exchanges between TSOs to their set-point. The tertiary control, activated manually and centrally at the TSO control centre, aims to restore the operating reserve, or to anticipate expected imbalances. Typically, the activation time is from 15 min up to several hours. 2.2. Frequency services in Denmark The Danish power system has the particular feature of belonging to both the Continental European synchronous area and the Nordic (Scandinavian) power system. Specifically, Western Denmark (DK1) which comprises Jutland peninsula and Funen island, has several AC connections to Germany and is therefore subject to the frequency services mentioned previously in Section 2.1. With reference to the primary frequency control, the mandatory share assigned to Western Denmark is equal to 27 MW. Eastern Denmark (DK2) instead has several AC connections to Sweden and is therefore synchronous to the Scandinavian Region. In the current framework, there is no automatic secondary frequency control in the Nordic power system— although a manual/tertiary reserve is in place and the primary frequency control is actually split in two separated services: frequency-controlled normal and frequency-controlled disturbance. The interested reader can get a more detailed overview on how the frequency control is organized in the Nordic system by consulting ref [24]. For what concerns the work described in this manuscript, it is interesting to analyse the technical conditions for providing the frequency-controlled disturbance service, which is actually rather demanding in term of response time: Supply inverse-linear power at frequencies between 49.9 and 49.5 Hz. Supply 50% of the response within 5 s. Supply the remaining 50% of the response within an additional 25 s. Accuracy and sensitivity of frequency measurements must be better than 10 mHz. SCADA system resolution must be better than 1 s. The listed requirements set the basis for benchmarking the performance of the experimental activity.
65
3. Controller characteristic and architecture and experimental layout 3.1. Primary frequency controller characteristics Commonly, primary frequency control is achieved via droop controllers, so that synchronous machines operating in parallel can share the load, according to their power rating. The droop constant is generally intended as the measure of the machine sensitivity to frequency changes, and is the value that quantifies its contribution to primary frequency/power regulation. The frequency variation, Df (in Hz), referred to the nominal frequency of the system is therefore given as a function of the relative power change DP (in W) or current change DI (in A) reported to the nominal machine power/current: . . . . aÞD f ¼ kdroop D P ; bÞD f ¼ kdroop D I ð1Þ f nom
Pnom
f nom
Inom
For example, a 5% droop means that a 5% frequency change (2.5 Hz) causes 100% change in the machine output. Defining a droop value for loads, on the other hand, may become less straightforward, as it may not be clear what the nominal power of the load – or of the set of loads – is. In this case, it has been considered that the available range of regulating power, that means 2.3 kW or 10 A per EV, is equal to the unit’s nominal power, instead of the nominal charging power of 3.7 kW or 16 A per EV. Therefore, in the experimental setup that will be described in Section 3.2, the following parameters are used for the EVs’ 4% droop control: 8 Df ¼ 2Hz; f nom ¼ 50Hz > > > > < DP ¼ 2:3kW; Pnom . ¼ 2:3kW Df ¼ 4%; aÞ nom > > .f kdroop ¼ > > DP : nom P
8 Df ¼ 2Hz; f nom ¼ 50Hz > > > > DI ¼ 10A; Inom < . ¼ 10A Df bÞ ¼ 4% nom > f > > > kdroop ¼ D . : I nom
ð2Þ
I
Fig. 1 reports the individual EV controlling characteristic, which means the current set-point in function of the frequency, for two different droops: 4% and 2%. The 2% droop is realized by simply reducing the delta frequency to 1 Hz (i.e., 49.5 50.5 Hz). It has to be reminded that the droop represent the slope of the curve if the plot reports the frequency in function of the current set-point. It is the inverse of the slope if the current is function of the frequency. The ideal droop curves, in dashed red and magenta in Fig. 1, spans respectively from 49.5 Hz to 50.5 Hz and 49 Hz to 51 Hz, and are linearly related to all the current values from 6 A to 16 A. The implemented ones, however, differ because the current values can be set with a discretization of 1 A. Therefore the ideal curve becomes quantized, and for every 0.05 Hz or 0.1 Hz, the current values are increased by 1 A. The minimum and maximum charging currents, respectively 6 A and 16 A, are maintained once the thresholds frequency are reached. This is the first main outcome, since the current standard IEC61851 only allows 1A-discrete modulation, while Energinet.dk requirements specify a linear relationship. In case an aggregator would realize such control strategy, it would be necessary to have a sufficient pool of cars that, once aggregated, show an equivalent linear response. A possible solution, which is left for further investigations, could be achieved by shifting the reported characteristics so that,
66
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
not allowed to inject power into the grid. The charging level can, however, be modulated between 6 and 16 A with steps of 1 A. Positive power means EV charging (i.e., consumption). Diesel genset equipped with a 60 kVA synchronous generator, capable of providing an active power output up to 48 kW. Positive power means generation. 15 kW 190 kWh Vanadium Redox Battery (VRB) storage system, equipped with an inverter capable of providing up to 15 kW and 12 kVAr. Positive power means battery charging, therefore consumption. 10 kW Aircon wind turbine equipped with full converter. Nominal wind speed 11 m/s, active stall power control. Positive power means generation. The Aircon is connected to the system only during the last set of experiments.
Fig. 1. EV ideal (dashed lines) and real (solid lines) control characteristic for 2 droop values (2% and 4%).
for instance, the current step between 6 A and 7 A is not realized for all the EVs at 49.00, but at 49.01 Hz for a certain set of EVs, at 49.02 Hz for another set and so on. Given the limited set of cars for the current experiments, the linearity requirement will not be fulfilled. The careful reader will also notice that both droop curves are slightly asymmetric (i.e., the 11 A value is set between 49.8 and 50.0 Hz and not between 49.9 and 50.1 HZ), giving therefore a bit more controlling capability for the over-frequency situations. As it will become clearer while presenting the experimental results (see Section 4.2), this choice was made in order to compensate for recurring undershooting phenomenon of approximately 0.5 A in the effective charging current of the EVs. 3.2. Experimental layout The experiments are executed using the hardware and ICT infrastructure of SYSLAB, which is a small-scale power system in the PowerLabDK platform. SYSLAB consists of a number of real power components interconnected by a three-phase 400 V AC power grid, and paralleled with communication and control nodes arranged in a dedicated network. The complete test bed is distributed (more than 1 km) over the Risø Campus of the Technical University of Denmark. The system may be connected to the local grid or can be islanded if desired [25]. The following components are used for the experiments, with their capabilities and operating points summarized in Table 1: Three controllable EVs (Nissan Leaf) each equipped with single phase 16 A (230 V) charger and 24 kWh lithium battery storage. The charger does not utilize any V2G capability, which means it is
The diesel genset is used to provide inertia to the system. The governor of the genset, however, is disabled in order to avoid automatic frequency control. Frequency events are triggered by changing the VRB set-points. The three EV inverters are equipped with the droop controllers described in Section 3.1, and rely on frequency measurements routed via wired network connection. It means that the frequency measurement is not used for feeding a local controller, but is processed remotely, and sent via network to the computer controlling the charging post. This aspect is extremely important because it is common practice for power plants to provide primary frequency control by using local frequency measurements. Moreover, as mentioned in Section 2.2, in order to participate in the frequency regulations, the usage of measurement devices with high accuracy (at least 10 mHz) and high sampling rate (at least 1 value every second) is required. This feasibility has already been proved in a gridconnected setup with just few seconds’ latency [14]. The experimental setup, including both power components and communication architecture, is presented in Fig. 2. Since all the components are 3-phase except for the EVs, it has been necessary to create an intermediate phase splitter, shown in Fig. 3. The phases are permuted cyclically so that each EV is supplied on a different phase via a standard Mennekes (IEC 62196 Type 2) connector. The operating set-points of different components are chosen in order to push the system to the limits. The EVs’ initial charging level is chosen so that there is room to increase and decrease the charge level equally. 3.3. Centralized frequency communication architecture The test setup used in this work consists of 3 electric vehicle supply equipment (EVSE), each connected to a different phase. The communication and control setup are shown in Fig. 4. The control aggregator is connected to each EVSE by Ethernet, using the MODBUS protocol, and is used to set a maximum EV charging current. The aggregator is also connected to a measurement unit, also by Ethernet, using the MODBUS protocol, providing local frequency and power measurements. Each EVSE is connected to the EV following the IEC 61851 standard.
Table 1 Units’ capabilities overview. Units
Capability
Single Electric Vehicle (single phase charger) Set of 3 EVs Vanadium Redox Battery Diesel generator Aircon wind turbine
6 16 A per phase (equal to 1.4 3.7 kW) 4 11 kW 15 15 kW; 12 12 kVAr 0 48 kW; 20 30 kVAr 10 kW @11 m/s
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
67
always disabled so that it is entirely up to the 3 EVs to control the system frequency. System performance is evaluated by checking that the system does not black out, and that the frequency is kept within the prescribed droop control values. Voltage values are also observed during the whole length of the transients. This analysis allows also the investigation of issues that may arise when dealing with the practical implementation of frequency response, such as: communication latency, robustness of service algorithm, power and frequency measurement inaccuracies, and coordination of more sources, such as more vehicles providing this service. The tests scenarios are reported below: a) Determination of diesel inertia and verification of lack of governor reaction (1 kW load variation and no frequency control). b) Load power steps (alternatively 3 kW VRB set-point). c) Wind-power balancing. Aircon is connected to the system, and EVs have to balance wind variability.
4.1. Determination of diesel inertia and lack of governor reaction
Fig. 2. Experimental setup: power components and electrical connections in red; communication architecture in black. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Fig. 3. 3-phase splitter: 3 Mennekes plugs used to connect each EV to a separate phase.
According to this standard, the EV listens to the communication line (called the Control Pilot line) from the EVSE, in the form of a PWM signal. The duty cycle of this signal indicates the maximum charging current limit the EV is allowed to draw. In this work currents from 6A to 16A with 1A steps are used, as indicated in Section 3.1. The frequency is measured using a DEIF MTR-3 meter connected to phase A. Measurements are polled every second with an accuracy of 10 mHz and a maximum latency of 0.2 s.
The first test intends to prove that the diesel governor is effectively disabled, and also to evaluate the amount of inertia of the genset, which is not given in the datasheet. A 1 kW power imbalance is created at the 60th second of the test, by increasing the VRB power consumption from 4 to 5 kW so that the frequency starts to decline. The power flows of the three components – diesel, VRB and EVs – are reported in the first plot of Fig. 5. EVs are not frequency responsive, but keep on consuming the pre-set maximum power. It can be seen that the diesel actually increases its electrical torque, so the generated power increases from 15.3 to 16.3 kW. It must be noted, however, that this increase in the electrical torque is not balanced by a corresponding increase in the mechanical torque, as would happen if the governor was enabled. This can be derived from the frequency trend shown in the second plot of Fig. 5. The frequency decreases till the diesel under-frequency protection, set to 47.5 Hz, opens the genset breaker, and the system blackouts. Relevant information which can be derived from this experiment is the system inertia, calculated by measuring the frequency drop over a certain amount of time, given a certain power imbalance. Since the inverters that control the VRB and the EV chargers do not have any virtual inertia feature, it can be assumed that all the system inertia belongs to the combined rotating masses of the diesel genset, which means the reciprocating machine, flywheel, and synchronous generator. According to the equation of motion indicated in (3a) and considering the parameters reported in Table 2, the system inertia can be calculated as reported in (4). Equation 3a is valid as long as the rotational speed, vm, does not differ too much from the nominal speed, vn. aÞDP ¼ Jvn
Dvm 2 pf ; bÞvm ¼ p Dt
ð3Þ
4. Experimental investigation and results discussion The experiments are intended to test the EVs’ ability to modulate the charge level, in order to control the system frequency in the case of load variations and wind power fluctuations. The load events include an alternate load-increase and load-decrease so that both over- and under-frequency dynamics can be studied. The amplitude of the load event is equal to 30% of the initial load value; i.e., 3 kW. In the last set of experiments, an attempt to balance the variable wind production is performed. The diesel governor is
J¼
DP Dt 1000 W 120 s ¼ 122 kg m2 ¼ vn p Df 157rad s 3:14 2 Hz
ð4Þ
Since the physical value may not provide an immediate indication, it may be useful to report the inertia in per unit: 2H ¼
2 Jðvn Þ2 122 kg m2 157rad s ¼ 50 s ¼ Snom 60000 VA
ð5Þ
68
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
Fig. 4. Communication and control setup.
reciprocating engine which needs proper balancing, in order to smooth out the natural pistons pulsations. 4.2. Load power steps
Fig. 5. First plot: Diesel, 3 EVs and VRB active power outputs. Second plot: frequency measured at Diesel power meter.
The 2H, or acceleration time constant, may seem rather large if compared to the typical values of conventional power plants, which generally span from 8 to 12 s [26]. It has to be kept in mind that this unit is designed for islanded operation, and has a
The second test is intended to assess whether the EVs can cope with large load variations, and to verify if their response follows the predefined droop characteristic illustrated in Fig. 1. The units’ power outputs and the frequency measured at the diesel breaker are reported in Fig. 6 for the 2% droop controller. The initial conditions consist in: VRB consuming 2 kW, EVs consuming 7.2 kW (10.5 A each EV @ 230 V), diesel genset producing 9.6 kW (20% nominal power) and the rest accounts for system losses. As it will be better highlighted afterwards, even though the EVs are pre-set to consume 11 A, they actually draw less current. According to some discussions with OEM, the reason for this undershooting resides in the fact that OEM wants to be sure that if the PWM signal is limiting the charging current, the EV does not cross the set threshold even for few seconds. It can be seen immediately in Fig. 6 that, despite the large load variations, which values around 30% of the initial load, the frequency is maintained between 49.5 and 50.5 Hz as expected from the 2% droop controlling characteristic. At the second 5400, when the first positive load step (VRB power from 2 kW to 5 kW) happens, the EVs quickly reduce their consumptions, and the frequency stabilises at around 49.4 Hz. The opposite situation is experienced at the second 5650, during the negative load step (VRB power from 2 kW to 1 kW), when the frequency stabilises at around 50.4 Hz.
Table 2 System values.
DP (power imbalance)
Df/Dt (frequency drop
p (number of pair poles)
vn (generator nominal rotating speed)
2
1500 rpm
in the time interval) 1000 W
2 Hz/120 s
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
Fig. 6. First plot: Diesel, 3 EVs and VRB active power outputs (Aircon is disconnected). Second plot: frequency measured.
Naturally, during the triggering events, the large system inertia plays a relevant role, which helps to slow down the frequency variations. As it can be seen from the equation of motion, the frequency variation depends on both the power variation (DP) and the inertia (J or 2H). Since it is not possible to change the system inertia, it has been decided to test the most challenging situation i.e., the largest power step which is equal to 3 kW in the VRB output. It has to be pointed out that the EV flexibility range, acting as the primary reserve available in the system, equals to 3450 W assuming the initial current set-point of 11 A per EV (that is 5 A per phase under the nominal voltage of 230 V). Frequency variations caused by a larger power steps, cannot be stabilized by the 3 EVs. As mentioned in Section 4.1, the system inertia is around 5 times larger (in term of relative values) compared to the one of national power systems. However, it has to be noted that the load variation under test is extremely large, being 30% of the initial load value. If, for example, the Continental European power system is taken into consideration, the sizing accident for primary frequency control is estimated in a power step of around 3 GW. If this number is compared with the maximum load value, which is roughly 520 GW, the normalized load variation is equal to 0.6%. One could argue that the diesel used in this experiment also operates in a lowloaded situation (i.e., 9.6 kW production for a 60 kVA machine), so the load variation analysed in the test (i.e., 3 kW) should be referred to the nominal power of the diesel (i.e., 60 kVA). In this case, the normalized load variation is around 5% (i.e., 3/60), which is still 8 times larger compared to the value reported for Continental Europe. Coming back to the transient behaviour of the system, it is interesting to observe in Fig. 7 that, when the load is equal to the initial conditions (i.e., VRB power at 2 kW), the frequency is oscillating around 50 Hz with a periodicity of around 6 s. This means an oscillation with a frequency of 0.16 Hz that belongs to the group of local plant mode instability [27]. This is a clear indication that the controlling action of the EVs is too intense, leading the EVs to resonate with the diesel genset, which is naturally not a desirable operation. The 4% droop characteristic mentioned in Section 3.1 is investigated as well in order to verify whether a less intense proportional action would have been better. The wind balancing results, reported in Section 4.3, will evaluate system performances with both 2% and 4% droop.
69
Fig. 7. First plot: Diesel, 3 EVs power and set-point and VRB active power outputs. Second plot: frequency measured.
Concerning the analysed load step test, it is also worthwhile focusing on the transient behaviour of the system right after the VRB power step. A 20-s time window is reported Fig. 8, where it is possible to notice the sudden frequency drop right after the 3 kW increase and the controlling action realised by the 3 EVs. It can be noted in the third plot that the delay between the set and the calculated current is always around 1 s, at least in the time window displayed, while the delay between the measured and the set current is between 0 and 1 s. It is possible to notice few times that the measured and the set current are apparently synchronized. This is due to a limitation of the logging system that saves
Fig. 8. First plot: Diesel, 3 EVs power and set-point and VRB active power outputs. Second plot: frequency measured. Third plot: calculated, set and measured current for each EV.
70
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
data every second making it difficult to appreciate faster dynamics: in this case, it is reasonable to assume that the EV response delay is within 0.5 s. However, it is possible to claim that most of the times the whole latency, starting from frequency measurement to EVs injected current into the grid, is between 2 and 3 s, which is certainly compatible with the requirements set by Energinet.dk for frequency controlled disturbance. An extensive and longer correlation analyses is conducted for the tests involving wind power balancing in order to support this finding. Another relevant point concerns the current undershooting: as mentioned at the beginning of this section, the EVs almost never draw the current set by the controller, but steadily remain few decimal of ampere below the set value. This is due to the fact that the applied standard IEC 61851 defines a current limitation rather than a precise current set-point; therefore EVs are not required to follow the value, but just to limit the current. The reader will also notice that the EV connected to phase C (red line in the third plot of Fig. 8) is sometimes slightly slower than the other two and a bit more conservative when it comes to limiting the current drawn. The reason for this behaviour can be explained by the fact that, even though all cars are of the same type, the third one is actually two years older; therefore, the reason for the smaller delay could reside in a better design of the EV charging system. It has to be mentioned that the standard IEC 61851 specifies that this delay should be within 3 s. 4.3. Wind power balancing scenarios In the final scenario, the effectiveness of EVs in balancing windpower production is tested. Both droop characteristics are investigated in a similar setup, even though it has been necessary to reduce the VRB set-point during the 4% droop trial due to
reduced wind speed. The initial values for both scenarios are summarized in Table 3. Two 20-min time series results are reported in Fig. 9 for the two scenarios. The first impression is that the frequency is much more stable with the 4% droop; however, it has to be mentioned that wind conditions changed during the test and the wind turbine produced less power during the 4% test. However, in authors’ opinion, this is not significantly impacting the controller’s performance. In fact, as it was shown in the previous subsection, the EVs were marginally stable when following the load steps even without a variable power source in the system. In any case, tuning the controller is not the most important test objective at this stage, but rather verifying that EVs are able to sustain the system, and assessing their response time. In order to get a more clear evaluation of the response time over the two tests, a correlation between phase A EV current and frequency measurements is reported. The left plot of Fig. 10 reports this correlation, highlighting also the applied control characteristic (red solid curve). The same correlation is moreover obtained by shifting the measured frequency respectively 1 s (middle plot) and 2 s (right plot) in the future. A detailed numerical correlation analysis is reported in Table 4, by considering time shifts between 0 and 4 s and three classes of correlations are evaluated. I calculated—I set: correlation between the calculated current set-point, which is based on the measured frequency and the current set by the controller. This operation is done in parallel for all the machines therefore the correlation is the same. It can be seen that the highest correlation (97.5%) is obtained for 1 s shift, meaning that it is reasonable to assume this kind of delay between 0.5 and 1.5 s.
Table 3 System values during the wind power balancing scenarios. Droop
Test duration
Diesel set-point (generation)
2% (6 16 A; 49.5 50.5 Hz) 4% (6 16 A; 49 51 Hz)
1800 s
9.6 kW (20% nominal power) 7 kW
1800 s
9.6 kW (20% nominal power) 4 kW
VRB set-point EV initial charging value (consumption) Expected wind power (generation) (consumption) 7.2 kW (11 A set per EV) 7.2 kW (11 A set per EV)
4 7 kW 2 5 kW
Fig. 9. First plot: Diesel, 3 EVs, VRB and Aircon active power outputs. Second plot: frequency. Dataset shown 20 min. EVs droop controller set to 2% (left plots) and to 4% (right plots).
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
71
Fig. 10. Correlation between Phase A EV current and frequency measurements. Left plot: no time shift; middle plot: 1 s time shift; right plot: 2 s time shift. 2% droop controller.
Table 4 Correlation between calculated/set/measured phase currents Droop 2% wind balancing data set 1800seconds.
I calculated—I set I set—I measured (ph A; ph B; ph C)
I calculated—I measured (ph A; ph B; ph C)
0 s shift
1 s shift
2 s shift
3 s shift
4 s shift
77.7% 91.7% 93.3% 62.6% 66.9% 65.9% 41.9%
97.5% 90.5% 86.0% 65.0% 90.7% 89.9% 59.0%
83.1% 67.4% 62.2% 53.6% 93.0% 93.5% 67.0%
57.4% 50.1% 47.8% 41.5% 70.0% 70.9% 58.2%
45.1% 53.9% 57.9% 40.9% 50.6% 51.0% 44.0%
I set—I measured: correlation between the calculated current set-point and the current effectively consumed (and measured) by the EV. The highest correlation can be found in the first column (0 s shift), although also the second column (1 s shift) presents quite high correlation values. It is reasonable to assume that the average response time is within 1 s. It has to be stressed that this correlation, lower than the previous one, considers also the mismatching between measured and set current due to the undershooting phenomena, highlighted in Section 4.2. It can be also appreciated that the 3rd EV is actually the one presenting the poorest performance compared to the other two.
I calculated—I measured: this correlation includes the two presented before and it is calculated in order to assess the overall EV response time. In this case the third column (2 s shift) is the one presenting the highest correlation, even though also the second one is rather large. It can be concluded that the overall response time (from frequency measurement to EV current measurement) is between 1.5 and 2.5 s. The same analysis is also performed for the 4% test case and the results are displayed in Fig. 11 and are reported analytically in
Fig. 11. Correlation between Phase A EV current and frequency measurements. Left plot: no time shift; middle plot: 1 s time shift; right plot: 2 s time shift. 4% droop controller.
72
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
Table 5 Correlation between calculated/set/measured phase currents Droop 4% wind balancing data set 1800seconds.
I calculated—I set I set—I measured (ph A; ph B; ph C)
I calculated—I measured (ph A; ph B; ph C)
0 s shift
1 s shift
2 s shift
3 s shift
4 s shift
67.4% 89.6% 92.1% 87.2% 66.0% 65.9% 64.6%
99.5% 80.6% 79.0% 85.6% 89.2% 91.2% 86.8%
67.4% 60.9% 61.7% 67.3% 80.5% 79.0% 85.4%
56.7% 54.2% 54.3% 58.7% 61.0% 61.8% 67.3%
50.4% 61.5% 62.9% 63.1% 54.6% 54.7% 59.0%
Table 5. The results are consistent with the previous ones and due to better performances of the controller, it can be appreciated a higher correlation for the “I calculated—I measured” with 1 s shift instead of the 2 s shift. Another important aspect relates to the quality of the frequency measurements. It is important to have a system robust against erroneous or missing information. In case faulty frequency measurements are detected, e.g. 0 Hz, the possibility of considering the last received frequency measurement as valid could be implemented, so EV control would be done according to that measurement. An alternative, although more expensive, could be the application of the 2 out of 3 logic often used in industrial processes. In this case it would be necessary for the aggregator to receive 3 different frequency measurements from separate instruments, possibly not electrically too far from each other, and then exclude the faulty one. In any case, similar approaches would still be cheaper compared than installing frequency meters in each EVSE. A final consideration relates to the possibility of applying the same control structure also for deploying secondary frequency control. In fact, the EV current limitation set-point could be based on an integral effect of the frequency error similarly to what is currently done, where set-points are received by dedicated power stations. From the aggregator perspective, it could be foreseen that a certain amount of EVs are allocated to modulate their power outputs in order to restore frequency errors (in case the secondary frequency control is used for restoring the frequency) or restoring power exchanges between TSO control areas. Alternatively, other energy resources, characterized by slower dynamics, could be used for providing secondary control. For instance, experimental results reported in [28] proved the provision of secondary frequency control by using thermostatic controllable loads, such as domestic refrigerators. In any case, it is important to properly size the amount of reserve devoted to primary and secondary frequency reserve, as well as the frequency activation ranges in order to avoid lack of regulating power when needed. 5. Conclusions and future activities The experimental results clearly show that frequency control from EVs is technically feasible with very fast response time. Therefore, coming back to the initial question, it is authors’ opinion that EVs can be a reliable source for this kind of service and that providing primary frequency control with centralized communication architecture, relying on decentralized energy resources is technically feasible. The testing proved that already today, with existing commercial vehicles and standards without any particular V2G capability, it is possible to achieve this objective. Naturally, there is room for improvement and a few points that the authors think are necessary to tackle are listed. Given the limited amount of regulating power, the chosen 2% droop value is rather small, meaning that the EVs’ controlling action is rather intense, which is one source of oscillations. The less intense 4% controlling action proved better transient performances. The two
new Leafs showed better performances, with the overall response delay always in the range of 2–3 s, while for the older one, the delay was sometimes a bit higher and the current undershooting slightly larger. Moreover, accuracy of the current limitation feature need to be better exploited in order to compensate the undershooting. The observed communication delay is limited; therefore deploying this kind of strategy via the Internet is not seen as an obstacle to this approach. The granularity of the control, with the 1 A discretization, does not help to obtain a smooth response, as each EV can change its output with a 10% step. Naturally, this response could become smoother in the moment that an aggregated set of EVs is considered (i.e., 10 or more). The controller is actually setting a pure current reference: future work will also investigate a voltage-compensation loop, in order to take into account considerable voltage deviations as well as compensating for larger than expected undershooting. As a final remark, it was interesting to appreciate standard EV chargers withstanding frequency events spanning from 47 up to 52 Hz without any interruption. Acknowledgments This work is supported by the Danish Research Project “NIKOLAIntelligent Electric Vehicle Integration”-under ForskEL kontrakt nr. 2013-1-12088. More information: www.nikolaproject. info. The authors are grateful to Nissan Denmark for providing 2 of the 3 Leafs used in the experiments. References [1] L. Martini, L. Radaelli, H. Brunner, C. Caerts, A. Morch, S. Hanninen, C. Tornelli, Electra IRP approach to voltage and frequency control for future power systems with high der penetration, 23rd International Conference on Electricity Distribution CIRED, Lyon, 15–18 June 2015, 2015, pp. 1–5. [2] K. Visscher, M. Marinelli, A.Z. Morch, S.H. Jakobsen, Identification of observables for future grids—The framework developed in the ELECTRA project, PowerTech, 2015 IEEE Eindhoven, June 29 2015–July 2 2015, 2016, pp. 1–6. [3] M. Rezkalla, K. Heussen, M. Marinelli, J. Hu, H.W. Bindner, Identification of requirements for distribution management systems in the smart grid context, Power Engineering Conference (UPEC), 50th International Universities, 1–4 Sept. 2015, 2015, pp. 1–6. [4] A. Zegers, H. Brunner, TSO-DSO interaction: an overview of current interaction between transmission and distribution system operators and an assessment of their cooperation in Smart Grids, ISGAN (International Smart Grid Action Network) Discussion Paper Annex 6 Power T&D Systems, Task 5, September, 2014. [5] R. Yan, T.K. Saha, N. Modi, N.-A. Masood, M. Mosadeghy, The combined effects of high penetration of wind and PV on power system frequency response, Appl. Energy 145 (2015) 320–330, doi:http://dx.doi.org/10.1016/j. apenergy.2015.02.044. [6] J. Fleer, P. Stenzel, Impact analysis of different operation strategies for battery energy storage systems providing primary control reserve, J. Energy Storage (2016), doi:http://dx.doi.org/10.1016/j.est.2016.02.003. [7] E. Kaempf, H. Abele, S. Stepanescu, M. Braun, Reactive power provision by distribution system operators—optimizing use of available flexibility, Innovative Smart Grid Technologies Conference Europe (ISGT-Europe), IEEE PES, 12–15 Oct. 2014, 2014, pp. 1–5. [8] J. García-Villalobos, I. Zamora, J.I. San Martín, F.J. Asensio, V. Aperribay, Plug-in electric vehicles in electric distribution networks: a review of smart charging
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
approaches, Renew. Sustain. Energy Rev. 38 (2014) 717–731, doi:http://dx.doi. org/10.1016/j.rser.2014.07.040. J. Hu, S. You, M. Lind, J. Østergaard, Coordinated charging of electric vehicles for congestion prevention in the distribution grid, IEEE Trans. Smart Grid 5 (2014) 703–711, doi:http://dx.doi.org/10.1109/TSG.2013.2279007. F. Mwasilu, J.J. Justo, E.K. Kim, T.D. Do, J.W. Jung, Electric vehicles and smart grid interaction: a review on vehicle to grid and renewable energy sources integration, Renew. Sustain. Energy Rev. 34 (2014) 501–516, doi:http://dx.doi. org/10.1016/j.rser.2014.03.031. K. Knezovi c, M. Marinelli, P. Codani, Y. Perez, Distribution grid services and flexibility provision by electric vehicles: a review of options, Power Engineering Conference (UPEC), 50th International Universities, 1–4 Sept. 2015, 2015, pp. 1–6. C. Gouveia, C.L. Moreira, J.A.P. Lopes, D. Varajao, R.E. Araujo, Microgrid service restoration: the role of plugged-in electric vehicles, IEEE Industrial Electronics Magazine, vol. 7, no. 4, Dec., 2013, pp. 26–41, doi:http://dx.doi.org/10.1109/ MIE.2013.2272337. W. Kempton, J. Tomi c, Vehicle-to-grid power implementation: from stabilizing the grid to supporting large-scale renewable energy, J. Power Sources 144 (2005) 280–294, doi:http://dx.doi.org/10.1016/j.jpowsour.2004.12.022. S. Martinenas, M. Marinelli, P.B. Andersen, C. Træholt, Implementation and demonstration of grid frequency support by V2G enabled electric vehicle, Proc 49th Int Univ Power Eng Conf IEEE (2014) 1–6, doi:http://dx.doi.org/10.1109/ UPEC.2014.6934760. S. Izadkhast, P. Garcia-Gonzalez, P. Frias, An aggregate model of plug-in electric vehicles for primary frequency control, IEEE Trans. Power Syst. 30 (2015) 1475–1482, doi:http://dx.doi.org/10.1109/TPWRS.2014.2337373. H. Liu, Z. Hu, Y. Song, J. Lin, Decentralized vehicle-to-grid control for primary frequency regulation considering charging demands, IEEE Trans. Power Syst. 28 (2013) 3480–3489, doi:http://dx.doi.org/10.1109/TPWRS.2013.2252029. P.M.R. Almeida, J.A.P. Lopes, F.J. Soares, L. Seca, Electric vehicles participating in frequency control: operating islanded systems with large penetration of renewable power sources, 2011 IEEE PES Trondheim PowerTech Power Technol a Sustain Soc POWERTECH 1 (2011) 5–10, doi:http://dx.doi.org/ 10.1109/PTC.2011.6019424.
73
[18] T. Masuta, A. Yokoyama, Supplementary load frequency control by use of a number of both electric vehicles and heat pump water heaters, IEEE Trans. Smart Grid 3 (2012) 1253–1262, doi:http://dx.doi.org/10.1109/ TSG.2012.2194746. [19] J. Meng, Y. Mu, H. Jia, J. Wu, X. Yu, B. Qu, Dynamic frequency response from electric vehicles considering travelling behavior in the Great Britain power system, Appl. Energy 162 (2016) 966–979, doi:http://dx.doi.org/10.1016/j. apenergy.2015.10.159. [20] J. Druitt, W.-G. Früh, Simulation of demand management and grid balancing with electric vehicles, J. Power Sources 216 (2012) 104–116, doi:http://dx.doi. org/10.1016/j.jpowsour.2012.05.033. [21] X. Luo, S. Xia, K.W. Chan, A decentralized charging control strategy for plug-in electric vehicles to mitigate wind farm intermittency and enhance frequency regulation, J. Power Sources 248 (2014) 604–614, doi:http://dx.doi.org/ 10.1016/j.jpowsour.2013.09.116. [22] P.B. Andersen, M. Marinelli, O.J. Olesen, C.A. Andersen, G. Poilasne, B. Christensen, et al., The Nikola project intelligent electric vehicle integration, IEEE PES Innov Smart Grid Technol Eur (2014) 1–6, doi:http://dx.doi.org/ 10.1109/ISGTEurope.2014.7028765. [23] ENTSO-E, Continental Europe Operation Handbook. Available: https://www. entsoe.eu/publications/system-operations-reports/operation-handbook/ Pages/default.aspx. [24] Energinet.dk, Ancillary services to be delivered in Denmark—Tender conditions. Available: https://energinet.dk/EN/Soeg/Sider/ Publikationsvisning.aspx?Filter1Value=4812&Filter1Field=Id. [25] M. Marinelli, F. Sossan, G.T. Costanzo, H.W. Bindner, Testing of a predictive control strategy for balancing renewable sources in a microgrid, Sustain. Energy IEEE Trans. 5 (2014) 1426–1433, doi:http://dx.doi.org/10.1109/ TSTE.2013.2294194. [26] P. Kundur, Power System Stability and Control, McGraw-Hill Education, 1994. [27] L.L. Grigsby, Power System Stability and Control. The Electric Power Engineering Handbook, CRC Press/IEEE Press, 2001, 2016. [28] V. Lakshmanan, M. Marinelli, J. Hu, H.W. Bindner, Provision of secondary frequency control via demand response activation on thermostatically controlled loads: solutions and experiences from Denmark, Appl. Energy 173 (2016) 470–480, doi:http://dx.doi.org/10.1016/j.apenergy.2016.04.054.
Contents lists available at ScienceDirect
Journal of Energy Storage journal homepage: www.elsevier.com/locate/est
Validating a centralized approach to primary frequency control with series-produced electric vehicles Mattia Marinelli* , Sergejus Martinenas, Katarina Knezovi c, Peter Bach Andersen Centre for Electric Power and Energy, Technical University of Denmark, Risø Campus, Roskilde, Denmark
A R T I C L E I N F O
Article history: Received 19 March 2016 Received in revised form 16 May 2016 Accepted 17 May 2016 Available online xxx Keywords: Centralized control Electric vehicles Frequency control Islanded systems Testing
A B S T R A C T
The aim of this work is twofold: on one hand it proposes a centralized approach to primary frequency control by using electric vehicles as controllable units; on the other hand, it experimentally validates whether series-produced EVs, adhering to contemporary standards, can be an effective resource for providing primary frequency control. The validation process is realized in an islanded system with renewable sources and it relies on verifying that the frequency values are within the desired limits following severe load steps or wind power fluctuations. In order to reflect today’s situation, the used EVs, three Nissan Leaf, are not taking advantage of any V2G capability, but rely solely on the possibility of limiting the charge between 6 A and 16 A. The centralized approach implies that the frequency is not measured locally as it is a common practice today, but is routed via the Internet in order to include potential communication delays that would take into account the presence of different entities for controlling the vehicles, such as aggregators and utilities. The centralised approach is pursued to support aggregators in participating in current ancillary service markets. Ultimately, this paper aims to strengthen the applied research within EV integration through the practical validation of smart grid concepts on original manufactured equipment. ã 2016 Elsevier Ltd. All rights reserved.
1. Introduction With conventional units being replaced by renewable resources, there is an increased demand for additional ancillary services, such as frequency control. Due to their defining property of being quick-response high-power units, electric vehicles (EVs) emerge as a viable actor for providing frequency regulation. A noticeable amount of research efforts is put nowadays to investigate all different aspects needed to shift from the traditional power system, where frequency and voltages are controlled by a relatively small set of large units, into a futuristic system, where potentially all power devices can be involved in the controlling
Abbreviations: DSO, distribution system operator; ENTSO-E, European Network of Transmission System Operators for Electricity; EV, electric vehicle; EVSE, electric vehicle supply equipment; FCR, frequency containment reserve; FRR, frequency restoration reserve; ICT, information and communications technology; OEM, original equipment manufacturer; OR, Operating reserve; PFC, primary frequency control; RES, renewable energy sources; RR, Replacement reserve; SOC, state-ofcharge; TSO, transmission system operator; V2G, vehicle to grid. * Corresponding author. E-mail addresses: [email protected], [email protected] (M. Marinelli). http://dx.doi.org/10.1016/j.est.2016.05.008 2352-152X/ã 2016 Elsevier Ltd. All rights reserved.
actions. For instance, in the ELECTRA project, innovative control schemes are being investigated in order to assess whether conventional frequency and voltage controlling approaches are still suitable, or which aspects need to be revised for scenarios with massive amounts of small distributed energy resources [1,2]. It is argued that TSOs alone will not be able to effectively manage the overall system balance and will have to partially delegate frequency control responsibility to DSOs as already happens today for congestions management and volt-reactive power provision [3–10]. Regardless of who will be responsible for frequency and voltage control, several technical challenges are ahead. For instance, will the aggregated provision of thousands of few kVA-units at low voltage level be as effective as the one of MVA-units response? Moreover, considering the traditional 3-phase system, will the response of groups of single-phase devices be as effective as the response of similar size 3-phase units? With a specific focus on frequency control, is it reasonable to expect millions of accurate local frequency measurements to be used for decentralized, traditional, droop controllers or is it rather better to have a limited set of centralized controllers, which rely on a few accurate
64
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
measurements, sending out power set-points via normal Internet connection? 1.1. Background and literature analysis Electric vehicles are one of the imminent candidates for providing ancillary services. Most of the time (typically about 90%) they are plugged into a charging post and can, in principle, provide fast-regulating power in both directions, or just modulate the charging power. A noticeable amount of literature has already been written supporting this statement [7–22]. It is argued that EVs with V2G capability can provide regulation services, and can compete in electricity markets, such as markets for ancillary services, where there is payment for available capacity, apart from the payment for the actual dispatch. Frequency control is one of the services which can be provided by EVs through this market. More specifically, primary frequency control (PFC) can be suitably provided by EVs due to their flexible operating mode and ability to seamlessly alter the consuming/ producing power under the V2G concept [12–14]. The work described in [15] presents an aggregated PFC model, where a participation factor, dependent on the state-of-charge (SOC), is associated with each EV to determine its droop characteristic. It was shown that EVs can effectively improve the system frequency response, as well as that V2G-capable vehicles have better power response, due to more available primary reserves. Furthermore, a decentralized V2G control for primary frequency regulation is presented in [16]. The proposed method considers customer charging demands and adapts the frequency droop control to maintain or achieve the desired SOC. A comparative study is performed in [17], in order to evaluate benefits of EVs performing primary frequency control in an islanded system with high penetration of renewable resources. The presented study case argues that system frequency oscillates in a 0.3 Hz band if the EVs contribute to primary regulation, compared to 1 Hz in the case of only using available hydro units. Two studies, respectively from Japan and Great Britain, analyse participation in frequency control on large systems considering traveling constraints and including large amount of renewable sources [18,19]. However, even though the mentioned studies analysed different strategies for providing primary frequency control, rarely have they dealt with the experimental validation, but mostly remained on modelling and simulations. For example, the works described in [15–21] have implemented different droop controls and shown that EVs can be effective in primary frequency regulation, likewise in isolated microgrids and larger systems. Still, they assume an ideal EV response to the control signals, both in the reaction time and the provided power, and they omit communication and control latencies which may greatly impact the results. In addition, technical challenges may arise due to the limited power and energy size of each individual unit, as well as the need to have simple and effective measuring and controlling capabilities for primary frequency regulation. Transmission systems operators may be sceptical about the possibility of having demand participating in the frequency regulation, mainly because of response uncertainties and metering inaccuracies. Therefore, an extensive experimental activity is required to prove the feasibility of these solutions. The activity described in this paper is carried out using series-produced vehicles and the universally supported IEC 61851 standard, to prove the applicability of the solution.
control, whereas the experimental validation is rarely touched upon. Therefore, this paper focuses on the evaluation of EVs’ ability to provide primary frequency control in a centralized fashion. The frequency control analysis of three EVs connected to a small islanded power system is proposed. Having the system islanded gives the possibility of emulating realistic frequency events, which could hardly be appreciated if connected to the national grid. It is important to note that the experiments are carried out with commercially available vehicles without taking advantage of any V2G capability, but only with the possibility to modulate the unidirectional charging current. This limit is part of internal standards (IEC61851/J1772) for conductive AC charging and supported by the vast majority of EVs today. A classical droop control function is utilized. However, contrary to today’s practice, which relies on local measurement, the frequency measurement is routed via the Internet to the controller, which sends the current set-point to the EV. The authors believe that in the near future, it will be highly unlikely to equip each EV with a precise measurement device which meets the TSO requirements. This means that the used technology resembles that of an operational environment: a fleet of EVs providing frequency regulation on market terms through an aggregator who has the certified frequency measurement device. Such setup allows the inclusion and assessment of potential communication delays, especially their influence on system stability. This work does not take in consideration vehicle unavailability due to owner usage. However it has to be reminded that, due to the system wide nature of the service, the location of the resource providing frequency control is not extremely important. In that sense, it is the aggregator’s best interest to rely on a higher number of vehicles to account for the plug-in uncertainties, whilst it is less important to know their exact location, as long as the utilized vehicles are connected to the grid and not used for other services. Ultimately, the research question tackled in this paper is: can small size, single phase distributed energy resources, such as commercially available electric vehicles, effectively provide primary frequency control relying on a centralized controller which sends out current set-points? The rest of the paper is structured as follows: Section 2 briefly recalls how the control of frequency is traditionally organized in Europe and in Denmark. Section 3 describes the controller characteristics, the communication architecture and the implementation in the laboratory. In Section 4 numerical and graphical results of the experiments are presented and discussed; several scenarios are investigated from load step, through steady state analysis to wind power balancing. Section 5 reports conclusions and lessons learned. 2. Current framework for frequency control in Europe and Denmark 2.1. Frequency services according to ENTSO-E division Based on the European Network of Transmission System Operators for Electricity (ENTSO-E) definitions, reported in the Network Code and Operation Handbook, frequency control includes [23]: Primary frequency control; Secondary power-frequency control; Tertiary control.
1.2. Objective of the manuscript Most of the literature identified during the literature review focuses on modeling and simulating the EV primary frequency
ENTSO-E refers to the reserves for frequency control as Operating Reserves (OR), and specifically, indicates the abovementioned controls as:
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
Frequency Containment Reserves (FCR); Frequency Restoration Reserves (FRR); Replacement Reserves (RR). The (automatic) primary frequency control aims at achieving the operational reliability of the synchronous area by stabilizing the system frequency after a disturbance or an incident at an acceptable stationary value in the second time frame, typically within 30 s. The minimum requirement for FCR is based on the maximum loss of power generation or demand, also called reference incident: within the Continental European synchronous area, the reference incident is defined as 3000 MW of generating capacity. FCR is shared among all control areas. The requirement of each control area is defined by contribution coefficients, calculated as the ratio between the energy generated in that area over a year and the energy generated in the entire synchronous area. The response has to be maintained for up to 15 min. The secondary frequency control, which can be either automatic or manual, aims to restore the system frequency within a few minutes, typically up to 15 min after incident, by releasing systemwide activated frequency containment reserves. For large interconnected systems, where a decentralized frequency restoration control is implemented, frequency restoration also aims at restoring the balance between generationand load foreach TSO, andconsequently, restores power exchanges between TSOs to their set-point. The tertiary control, activated manually and centrally at the TSO control centre, aims to restore the operating reserve, or to anticipate expected imbalances. Typically, the activation time is from 15 min up to several hours. 2.2. Frequency services in Denmark The Danish power system has the particular feature of belonging to both the Continental European synchronous area and the Nordic (Scandinavian) power system. Specifically, Western Denmark (DK1) which comprises Jutland peninsula and Funen island, has several AC connections to Germany and is therefore subject to the frequency services mentioned previously in Section 2.1. With reference to the primary frequency control, the mandatory share assigned to Western Denmark is equal to 27 MW. Eastern Denmark (DK2) instead has several AC connections to Sweden and is therefore synchronous to the Scandinavian Region. In the current framework, there is no automatic secondary frequency control in the Nordic power system— although a manual/tertiary reserve is in place and the primary frequency control is actually split in two separated services: frequency-controlled normal and frequency-controlled disturbance. The interested reader can get a more detailed overview on how the frequency control is organized in the Nordic system by consulting ref [24]. For what concerns the work described in this manuscript, it is interesting to analyse the technical conditions for providing the frequency-controlled disturbance service, which is actually rather demanding in term of response time: Supply inverse-linear power at frequencies between 49.9 and 49.5 Hz. Supply 50% of the response within 5 s. Supply the remaining 50% of the response within an additional 25 s. Accuracy and sensitivity of frequency measurements must be better than 10 mHz. SCADA system resolution must be better than 1 s. The listed requirements set the basis for benchmarking the performance of the experimental activity.
65
3. Controller characteristic and architecture and experimental layout 3.1. Primary frequency controller characteristics Commonly, primary frequency control is achieved via droop controllers, so that synchronous machines operating in parallel can share the load, according to their power rating. The droop constant is generally intended as the measure of the machine sensitivity to frequency changes, and is the value that quantifies its contribution to primary frequency/power regulation. The frequency variation, Df (in Hz), referred to the nominal frequency of the system is therefore given as a function of the relative power change DP (in W) or current change DI (in A) reported to the nominal machine power/current: . . . . aÞD f ¼ kdroop D P ; bÞD f ¼ kdroop D I ð1Þ f nom
Pnom
f nom
Inom
For example, a 5% droop means that a 5% frequency change (2.5 Hz) causes 100% change in the machine output. Defining a droop value for loads, on the other hand, may become less straightforward, as it may not be clear what the nominal power of the load – or of the set of loads – is. In this case, it has been considered that the available range of regulating power, that means 2.3 kW or 10 A per EV, is equal to the unit’s nominal power, instead of the nominal charging power of 3.7 kW or 16 A per EV. Therefore, in the experimental setup that will be described in Section 3.2, the following parameters are used for the EVs’ 4% droop control: 8 Df ¼ 2Hz; f nom ¼ 50Hz > > > > < DP ¼ 2:3kW; Pnom . ¼ 2:3kW Df ¼ 4%; aÞ nom > > .f kdroop ¼ > > DP : nom P
8 Df ¼ 2Hz; f nom ¼ 50Hz > > > > DI ¼ 10A; Inom < . ¼ 10A Df bÞ ¼ 4% nom > f > > > kdroop ¼ D . : I nom
ð2Þ
I
Fig. 1 reports the individual EV controlling characteristic, which means the current set-point in function of the frequency, for two different droops: 4% and 2%. The 2% droop is realized by simply reducing the delta frequency to 1 Hz (i.e., 49.5 50.5 Hz). It has to be reminded that the droop represent the slope of the curve if the plot reports the frequency in function of the current set-point. It is the inverse of the slope if the current is function of the frequency. The ideal droop curves, in dashed red and magenta in Fig. 1, spans respectively from 49.5 Hz to 50.5 Hz and 49 Hz to 51 Hz, and are linearly related to all the current values from 6 A to 16 A. The implemented ones, however, differ because the current values can be set with a discretization of 1 A. Therefore the ideal curve becomes quantized, and for every 0.05 Hz or 0.1 Hz, the current values are increased by 1 A. The minimum and maximum charging currents, respectively 6 A and 16 A, are maintained once the thresholds frequency are reached. This is the first main outcome, since the current standard IEC61851 only allows 1A-discrete modulation, while Energinet.dk requirements specify a linear relationship. In case an aggregator would realize such control strategy, it would be necessary to have a sufficient pool of cars that, once aggregated, show an equivalent linear response. A possible solution, which is left for further investigations, could be achieved by shifting the reported characteristics so that,
66
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
not allowed to inject power into the grid. The charging level can, however, be modulated between 6 and 16 A with steps of 1 A. Positive power means EV charging (i.e., consumption). Diesel genset equipped with a 60 kVA synchronous generator, capable of providing an active power output up to 48 kW. Positive power means generation. 15 kW 190 kWh Vanadium Redox Battery (VRB) storage system, equipped with an inverter capable of providing up to 15 kW and 12 kVAr. Positive power means battery charging, therefore consumption. 10 kW Aircon wind turbine equipped with full converter. Nominal wind speed 11 m/s, active stall power control. Positive power means generation. The Aircon is connected to the system only during the last set of experiments.
Fig. 1. EV ideal (dashed lines) and real (solid lines) control characteristic for 2 droop values (2% and 4%).
for instance, the current step between 6 A and 7 A is not realized for all the EVs at 49.00, but at 49.01 Hz for a certain set of EVs, at 49.02 Hz for another set and so on. Given the limited set of cars for the current experiments, the linearity requirement will not be fulfilled. The careful reader will also notice that both droop curves are slightly asymmetric (i.e., the 11 A value is set between 49.8 and 50.0 Hz and not between 49.9 and 50.1 HZ), giving therefore a bit more controlling capability for the over-frequency situations. As it will become clearer while presenting the experimental results (see Section 4.2), this choice was made in order to compensate for recurring undershooting phenomenon of approximately 0.5 A in the effective charging current of the EVs. 3.2. Experimental layout The experiments are executed using the hardware and ICT infrastructure of SYSLAB, which is a small-scale power system in the PowerLabDK platform. SYSLAB consists of a number of real power components interconnected by a three-phase 400 V AC power grid, and paralleled with communication and control nodes arranged in a dedicated network. The complete test bed is distributed (more than 1 km) over the Risø Campus of the Technical University of Denmark. The system may be connected to the local grid or can be islanded if desired [25]. The following components are used for the experiments, with their capabilities and operating points summarized in Table 1: Three controllable EVs (Nissan Leaf) each equipped with single phase 16 A (230 V) charger and 24 kWh lithium battery storage. The charger does not utilize any V2G capability, which means it is
The diesel genset is used to provide inertia to the system. The governor of the genset, however, is disabled in order to avoid automatic frequency control. Frequency events are triggered by changing the VRB set-points. The three EV inverters are equipped with the droop controllers described in Section 3.1, and rely on frequency measurements routed via wired network connection. It means that the frequency measurement is not used for feeding a local controller, but is processed remotely, and sent via network to the computer controlling the charging post. This aspect is extremely important because it is common practice for power plants to provide primary frequency control by using local frequency measurements. Moreover, as mentioned in Section 2.2, in order to participate in the frequency regulations, the usage of measurement devices with high accuracy (at least 10 mHz) and high sampling rate (at least 1 value every second) is required. This feasibility has already been proved in a gridconnected setup with just few seconds’ latency [14]. The experimental setup, including both power components and communication architecture, is presented in Fig. 2. Since all the components are 3-phase except for the EVs, it has been necessary to create an intermediate phase splitter, shown in Fig. 3. The phases are permuted cyclically so that each EV is supplied on a different phase via a standard Mennekes (IEC 62196 Type 2) connector. The operating set-points of different components are chosen in order to push the system to the limits. The EVs’ initial charging level is chosen so that there is room to increase and decrease the charge level equally. 3.3. Centralized frequency communication architecture The test setup used in this work consists of 3 electric vehicle supply equipment (EVSE), each connected to a different phase. The communication and control setup are shown in Fig. 4. The control aggregator is connected to each EVSE by Ethernet, using the MODBUS protocol, and is used to set a maximum EV charging current. The aggregator is also connected to a measurement unit, also by Ethernet, using the MODBUS protocol, providing local frequency and power measurements. Each EVSE is connected to the EV following the IEC 61851 standard.
Table 1 Units’ capabilities overview. Units
Capability
Single Electric Vehicle (single phase charger) Set of 3 EVs Vanadium Redox Battery Diesel generator Aircon wind turbine
6 16 A per phase (equal to 1.4 3.7 kW) 4 11 kW 15 15 kW; 12 12 kVAr 0 48 kW; 20 30 kVAr 10 kW @11 m/s
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
67
always disabled so that it is entirely up to the 3 EVs to control the system frequency. System performance is evaluated by checking that the system does not black out, and that the frequency is kept within the prescribed droop control values. Voltage values are also observed during the whole length of the transients. This analysis allows also the investigation of issues that may arise when dealing with the practical implementation of frequency response, such as: communication latency, robustness of service algorithm, power and frequency measurement inaccuracies, and coordination of more sources, such as more vehicles providing this service. The tests scenarios are reported below: a) Determination of diesel inertia and verification of lack of governor reaction (1 kW load variation and no frequency control). b) Load power steps (alternatively 3 kW VRB set-point). c) Wind-power balancing. Aircon is connected to the system, and EVs have to balance wind variability.
4.1. Determination of diesel inertia and lack of governor reaction
Fig. 2. Experimental setup: power components and electrical connections in red; communication architecture in black. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Fig. 3. 3-phase splitter: 3 Mennekes plugs used to connect each EV to a separate phase.
According to this standard, the EV listens to the communication line (called the Control Pilot line) from the EVSE, in the form of a PWM signal. The duty cycle of this signal indicates the maximum charging current limit the EV is allowed to draw. In this work currents from 6A to 16A with 1A steps are used, as indicated in Section 3.1. The frequency is measured using a DEIF MTR-3 meter connected to phase A. Measurements are polled every second with an accuracy of 10 mHz and a maximum latency of 0.2 s.
The first test intends to prove that the diesel governor is effectively disabled, and also to evaluate the amount of inertia of the genset, which is not given in the datasheet. A 1 kW power imbalance is created at the 60th second of the test, by increasing the VRB power consumption from 4 to 5 kW so that the frequency starts to decline. The power flows of the three components – diesel, VRB and EVs – are reported in the first plot of Fig. 5. EVs are not frequency responsive, but keep on consuming the pre-set maximum power. It can be seen that the diesel actually increases its electrical torque, so the generated power increases from 15.3 to 16.3 kW. It must be noted, however, that this increase in the electrical torque is not balanced by a corresponding increase in the mechanical torque, as would happen if the governor was enabled. This can be derived from the frequency trend shown in the second plot of Fig. 5. The frequency decreases till the diesel under-frequency protection, set to 47.5 Hz, opens the genset breaker, and the system blackouts. Relevant information which can be derived from this experiment is the system inertia, calculated by measuring the frequency drop over a certain amount of time, given a certain power imbalance. Since the inverters that control the VRB and the EV chargers do not have any virtual inertia feature, it can be assumed that all the system inertia belongs to the combined rotating masses of the diesel genset, which means the reciprocating machine, flywheel, and synchronous generator. According to the equation of motion indicated in (3a) and considering the parameters reported in Table 2, the system inertia can be calculated as reported in (4). Equation 3a is valid as long as the rotational speed, vm, does not differ too much from the nominal speed, vn. aÞDP ¼ Jvn
Dvm 2 pf ; bÞvm ¼ p Dt
ð3Þ
4. Experimental investigation and results discussion The experiments are intended to test the EVs’ ability to modulate the charge level, in order to control the system frequency in the case of load variations and wind power fluctuations. The load events include an alternate load-increase and load-decrease so that both over- and under-frequency dynamics can be studied. The amplitude of the load event is equal to 30% of the initial load value; i.e., 3 kW. In the last set of experiments, an attempt to balance the variable wind production is performed. The diesel governor is
J¼
DP Dt 1000 W 120 s ¼ 122 kg m2 ¼ vn p Df 157rad s 3:14 2 Hz
ð4Þ
Since the physical value may not provide an immediate indication, it may be useful to report the inertia in per unit: 2H ¼
2 Jðvn Þ2 122 kg m2 157rad s ¼ 50 s ¼ Snom 60000 VA
ð5Þ
68
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
Fig. 4. Communication and control setup.
reciprocating engine which needs proper balancing, in order to smooth out the natural pistons pulsations. 4.2. Load power steps
Fig. 5. First plot: Diesel, 3 EVs and VRB active power outputs. Second plot: frequency measured at Diesel power meter.
The 2H, or acceleration time constant, may seem rather large if compared to the typical values of conventional power plants, which generally span from 8 to 12 s [26]. It has to be kept in mind that this unit is designed for islanded operation, and has a
The second test is intended to assess whether the EVs can cope with large load variations, and to verify if their response follows the predefined droop characteristic illustrated in Fig. 1. The units’ power outputs and the frequency measured at the diesel breaker are reported in Fig. 6 for the 2% droop controller. The initial conditions consist in: VRB consuming 2 kW, EVs consuming 7.2 kW (10.5 A each EV @ 230 V), diesel genset producing 9.6 kW (20% nominal power) and the rest accounts for system losses. As it will be better highlighted afterwards, even though the EVs are pre-set to consume 11 A, they actually draw less current. According to some discussions with OEM, the reason for this undershooting resides in the fact that OEM wants to be sure that if the PWM signal is limiting the charging current, the EV does not cross the set threshold even for few seconds. It can be seen immediately in Fig. 6 that, despite the large load variations, which values around 30% of the initial load, the frequency is maintained between 49.5 and 50.5 Hz as expected from the 2% droop controlling characteristic. At the second 5400, when the first positive load step (VRB power from 2 kW to 5 kW) happens, the EVs quickly reduce their consumptions, and the frequency stabilises at around 49.4 Hz. The opposite situation is experienced at the second 5650, during the negative load step (VRB power from 2 kW to 1 kW), when the frequency stabilises at around 50.4 Hz.
Table 2 System values.
DP (power imbalance)
Df/Dt (frequency drop
p (number of pair poles)
vn (generator nominal rotating speed)
2
1500 rpm
in the time interval) 1000 W
2 Hz/120 s
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
Fig. 6. First plot: Diesel, 3 EVs and VRB active power outputs (Aircon is disconnected). Second plot: frequency measured.
Naturally, during the triggering events, the large system inertia plays a relevant role, which helps to slow down the frequency variations. As it can be seen from the equation of motion, the frequency variation depends on both the power variation (DP) and the inertia (J or 2H). Since it is not possible to change the system inertia, it has been decided to test the most challenging situation i.e., the largest power step which is equal to 3 kW in the VRB output. It has to be pointed out that the EV flexibility range, acting as the primary reserve available in the system, equals to 3450 W assuming the initial current set-point of 11 A per EV (that is 5 A per phase under the nominal voltage of 230 V). Frequency variations caused by a larger power steps, cannot be stabilized by the 3 EVs. As mentioned in Section 4.1, the system inertia is around 5 times larger (in term of relative values) compared to the one of national power systems. However, it has to be noted that the load variation under test is extremely large, being 30% of the initial load value. If, for example, the Continental European power system is taken into consideration, the sizing accident for primary frequency control is estimated in a power step of around 3 GW. If this number is compared with the maximum load value, which is roughly 520 GW, the normalized load variation is equal to 0.6%. One could argue that the diesel used in this experiment also operates in a lowloaded situation (i.e., 9.6 kW production for a 60 kVA machine), so the load variation analysed in the test (i.e., 3 kW) should be referred to the nominal power of the diesel (i.e., 60 kVA). In this case, the normalized load variation is around 5% (i.e., 3/60), which is still 8 times larger compared to the value reported for Continental Europe. Coming back to the transient behaviour of the system, it is interesting to observe in Fig. 7 that, when the load is equal to the initial conditions (i.e., VRB power at 2 kW), the frequency is oscillating around 50 Hz with a periodicity of around 6 s. This means an oscillation with a frequency of 0.16 Hz that belongs to the group of local plant mode instability [27]. This is a clear indication that the controlling action of the EVs is too intense, leading the EVs to resonate with the diesel genset, which is naturally not a desirable operation. The 4% droop characteristic mentioned in Section 3.1 is investigated as well in order to verify whether a less intense proportional action would have been better. The wind balancing results, reported in Section 4.3, will evaluate system performances with both 2% and 4% droop.
69
Fig. 7. First plot: Diesel, 3 EVs power and set-point and VRB active power outputs. Second plot: frequency measured.
Concerning the analysed load step test, it is also worthwhile focusing on the transient behaviour of the system right after the VRB power step. A 20-s time window is reported Fig. 8, where it is possible to notice the sudden frequency drop right after the 3 kW increase and the controlling action realised by the 3 EVs. It can be noted in the third plot that the delay between the set and the calculated current is always around 1 s, at least in the time window displayed, while the delay between the measured and the set current is between 0 and 1 s. It is possible to notice few times that the measured and the set current are apparently synchronized. This is due to a limitation of the logging system that saves
Fig. 8. First plot: Diesel, 3 EVs power and set-point and VRB active power outputs. Second plot: frequency measured. Third plot: calculated, set and measured current for each EV.
70
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
data every second making it difficult to appreciate faster dynamics: in this case, it is reasonable to assume that the EV response delay is within 0.5 s. However, it is possible to claim that most of the times the whole latency, starting from frequency measurement to EVs injected current into the grid, is between 2 and 3 s, which is certainly compatible with the requirements set by Energinet.dk for frequency controlled disturbance. An extensive and longer correlation analyses is conducted for the tests involving wind power balancing in order to support this finding. Another relevant point concerns the current undershooting: as mentioned at the beginning of this section, the EVs almost never draw the current set by the controller, but steadily remain few decimal of ampere below the set value. This is due to the fact that the applied standard IEC 61851 defines a current limitation rather than a precise current set-point; therefore EVs are not required to follow the value, but just to limit the current. The reader will also notice that the EV connected to phase C (red line in the third plot of Fig. 8) is sometimes slightly slower than the other two and a bit more conservative when it comes to limiting the current drawn. The reason for this behaviour can be explained by the fact that, even though all cars are of the same type, the third one is actually two years older; therefore, the reason for the smaller delay could reside in a better design of the EV charging system. It has to be mentioned that the standard IEC 61851 specifies that this delay should be within 3 s. 4.3. Wind power balancing scenarios In the final scenario, the effectiveness of EVs in balancing windpower production is tested. Both droop characteristics are investigated in a similar setup, even though it has been necessary to reduce the VRB set-point during the 4% droop trial due to
reduced wind speed. The initial values for both scenarios are summarized in Table 3. Two 20-min time series results are reported in Fig. 9 for the two scenarios. The first impression is that the frequency is much more stable with the 4% droop; however, it has to be mentioned that wind conditions changed during the test and the wind turbine produced less power during the 4% test. However, in authors’ opinion, this is not significantly impacting the controller’s performance. In fact, as it was shown in the previous subsection, the EVs were marginally stable when following the load steps even without a variable power source in the system. In any case, tuning the controller is not the most important test objective at this stage, but rather verifying that EVs are able to sustain the system, and assessing their response time. In order to get a more clear evaluation of the response time over the two tests, a correlation between phase A EV current and frequency measurements is reported. The left plot of Fig. 10 reports this correlation, highlighting also the applied control characteristic (red solid curve). The same correlation is moreover obtained by shifting the measured frequency respectively 1 s (middle plot) and 2 s (right plot) in the future. A detailed numerical correlation analysis is reported in Table 4, by considering time shifts between 0 and 4 s and three classes of correlations are evaluated. I calculated—I set: correlation between the calculated current set-point, which is based on the measured frequency and the current set by the controller. This operation is done in parallel for all the machines therefore the correlation is the same. It can be seen that the highest correlation (97.5%) is obtained for 1 s shift, meaning that it is reasonable to assume this kind of delay between 0.5 and 1.5 s.
Table 3 System values during the wind power balancing scenarios. Droop
Test duration
Diesel set-point (generation)
2% (6 16 A; 49.5 50.5 Hz) 4% (6 16 A; 49 51 Hz)
1800 s
9.6 kW (20% nominal power) 7 kW
1800 s
9.6 kW (20% nominal power) 4 kW
VRB set-point EV initial charging value (consumption) Expected wind power (generation) (consumption) 7.2 kW (11 A set per EV) 7.2 kW (11 A set per EV)
4 7 kW 2 5 kW
Fig. 9. First plot: Diesel, 3 EVs, VRB and Aircon active power outputs. Second plot: frequency. Dataset shown 20 min. EVs droop controller set to 2% (left plots) and to 4% (right plots).
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
71
Fig. 10. Correlation between Phase A EV current and frequency measurements. Left plot: no time shift; middle plot: 1 s time shift; right plot: 2 s time shift. 2% droop controller.
Table 4 Correlation between calculated/set/measured phase currents Droop 2% wind balancing data set 1800seconds.
I calculated—I set I set—I measured (ph A; ph B; ph C)
I calculated—I measured (ph A; ph B; ph C)
0 s shift
1 s shift
2 s shift
3 s shift
4 s shift
77.7% 91.7% 93.3% 62.6% 66.9% 65.9% 41.9%
97.5% 90.5% 86.0% 65.0% 90.7% 89.9% 59.0%
83.1% 67.4% 62.2% 53.6% 93.0% 93.5% 67.0%
57.4% 50.1% 47.8% 41.5% 70.0% 70.9% 58.2%
45.1% 53.9% 57.9% 40.9% 50.6% 51.0% 44.0%
I set—I measured: correlation between the calculated current set-point and the current effectively consumed (and measured) by the EV. The highest correlation can be found in the first column (0 s shift), although also the second column (1 s shift) presents quite high correlation values. It is reasonable to assume that the average response time is within 1 s. It has to be stressed that this correlation, lower than the previous one, considers also the mismatching between measured and set current due to the undershooting phenomena, highlighted in Section 4.2. It can be also appreciated that the 3rd EV is actually the one presenting the poorest performance compared to the other two.
I calculated—I measured: this correlation includes the two presented before and it is calculated in order to assess the overall EV response time. In this case the third column (2 s shift) is the one presenting the highest correlation, even though also the second one is rather large. It can be concluded that the overall response time (from frequency measurement to EV current measurement) is between 1.5 and 2.5 s. The same analysis is also performed for the 4% test case and the results are displayed in Fig. 11 and are reported analytically in
Fig. 11. Correlation between Phase A EV current and frequency measurements. Left plot: no time shift; middle plot: 1 s time shift; right plot: 2 s time shift. 4% droop controller.
72
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
Table 5 Correlation between calculated/set/measured phase currents Droop 4% wind balancing data set 1800seconds.
I calculated—I set I set—I measured (ph A; ph B; ph C)
I calculated—I measured (ph A; ph B; ph C)
0 s shift
1 s shift
2 s shift
3 s shift
4 s shift
67.4% 89.6% 92.1% 87.2% 66.0% 65.9% 64.6%
99.5% 80.6% 79.0% 85.6% 89.2% 91.2% 86.8%
67.4% 60.9% 61.7% 67.3% 80.5% 79.0% 85.4%
56.7% 54.2% 54.3% 58.7% 61.0% 61.8% 67.3%
50.4% 61.5% 62.9% 63.1% 54.6% 54.7% 59.0%
Table 5. The results are consistent with the previous ones and due to better performances of the controller, it can be appreciated a higher correlation for the “I calculated—I measured” with 1 s shift instead of the 2 s shift. Another important aspect relates to the quality of the frequency measurements. It is important to have a system robust against erroneous or missing information. In case faulty frequency measurements are detected, e.g. 0 Hz, the possibility of considering the last received frequency measurement as valid could be implemented, so EV control would be done according to that measurement. An alternative, although more expensive, could be the application of the 2 out of 3 logic often used in industrial processes. In this case it would be necessary for the aggregator to receive 3 different frequency measurements from separate instruments, possibly not electrically too far from each other, and then exclude the faulty one. In any case, similar approaches would still be cheaper compared than installing frequency meters in each EVSE. A final consideration relates to the possibility of applying the same control structure also for deploying secondary frequency control. In fact, the EV current limitation set-point could be based on an integral effect of the frequency error similarly to what is currently done, where set-points are received by dedicated power stations. From the aggregator perspective, it could be foreseen that a certain amount of EVs are allocated to modulate their power outputs in order to restore frequency errors (in case the secondary frequency control is used for restoring the frequency) or restoring power exchanges between TSO control areas. Alternatively, other energy resources, characterized by slower dynamics, could be used for providing secondary control. For instance, experimental results reported in [28] proved the provision of secondary frequency control by using thermostatic controllable loads, such as domestic refrigerators. In any case, it is important to properly size the amount of reserve devoted to primary and secondary frequency reserve, as well as the frequency activation ranges in order to avoid lack of regulating power when needed. 5. Conclusions and future activities The experimental results clearly show that frequency control from EVs is technically feasible with very fast response time. Therefore, coming back to the initial question, it is authors’ opinion that EVs can be a reliable source for this kind of service and that providing primary frequency control with centralized communication architecture, relying on decentralized energy resources is technically feasible. The testing proved that already today, with existing commercial vehicles and standards without any particular V2G capability, it is possible to achieve this objective. Naturally, there is room for improvement and a few points that the authors think are necessary to tackle are listed. Given the limited amount of regulating power, the chosen 2% droop value is rather small, meaning that the EVs’ controlling action is rather intense, which is one source of oscillations. The less intense 4% controlling action proved better transient performances. The two
new Leafs showed better performances, with the overall response delay always in the range of 2–3 s, while for the older one, the delay was sometimes a bit higher and the current undershooting slightly larger. Moreover, accuracy of the current limitation feature need to be better exploited in order to compensate the undershooting. The observed communication delay is limited; therefore deploying this kind of strategy via the Internet is not seen as an obstacle to this approach. The granularity of the control, with the 1 A discretization, does not help to obtain a smooth response, as each EV can change its output with a 10% step. Naturally, this response could become smoother in the moment that an aggregated set of EVs is considered (i.e., 10 or more). The controller is actually setting a pure current reference: future work will also investigate a voltage-compensation loop, in order to take into account considerable voltage deviations as well as compensating for larger than expected undershooting. As a final remark, it was interesting to appreciate standard EV chargers withstanding frequency events spanning from 47 up to 52 Hz without any interruption. Acknowledgments This work is supported by the Danish Research Project “NIKOLAIntelligent Electric Vehicle Integration”-under ForskEL kontrakt nr. 2013-1-12088. More information: www.nikolaproject. info. The authors are grateful to Nissan Denmark for providing 2 of the 3 Leafs used in the experiments. References [1] L. Martini, L. Radaelli, H. Brunner, C. Caerts, A. Morch, S. Hanninen, C. Tornelli, Electra IRP approach to voltage and frequency control for future power systems with high der penetration, 23rd International Conference on Electricity Distribution CIRED, Lyon, 15–18 June 2015, 2015, pp. 1–5. [2] K. Visscher, M. Marinelli, A.Z. Morch, S.H. Jakobsen, Identification of observables for future grids—The framework developed in the ELECTRA project, PowerTech, 2015 IEEE Eindhoven, June 29 2015–July 2 2015, 2016, pp. 1–6. [3] M. Rezkalla, K. Heussen, M. Marinelli, J. Hu, H.W. Bindner, Identification of requirements for distribution management systems in the smart grid context, Power Engineering Conference (UPEC), 50th International Universities, 1–4 Sept. 2015, 2015, pp. 1–6. [4] A. Zegers, H. Brunner, TSO-DSO interaction: an overview of current interaction between transmission and distribution system operators and an assessment of their cooperation in Smart Grids, ISGAN (International Smart Grid Action Network) Discussion Paper Annex 6 Power T&D Systems, Task 5, September, 2014. [5] R. Yan, T.K. Saha, N. Modi, N.-A. Masood, M. Mosadeghy, The combined effects of high penetration of wind and PV on power system frequency response, Appl. Energy 145 (2015) 320–330, doi:http://dx.doi.org/10.1016/j. apenergy.2015.02.044. [6] J. Fleer, P. Stenzel, Impact analysis of different operation strategies for battery energy storage systems providing primary control reserve, J. Energy Storage (2016), doi:http://dx.doi.org/10.1016/j.est.2016.02.003. [7] E. Kaempf, H. Abele, S. Stepanescu, M. Braun, Reactive power provision by distribution system operators—optimizing use of available flexibility, Innovative Smart Grid Technologies Conference Europe (ISGT-Europe), IEEE PES, 12–15 Oct. 2014, 2014, pp. 1–5. [8] J. García-Villalobos, I. Zamora, J.I. San Martín, F.J. Asensio, V. Aperribay, Plug-in electric vehicles in electric distribution networks: a review of smart charging
M. Marinelli et al. / Journal of Energy Storage 7 (2016) 63–73
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
approaches, Renew. Sustain. Energy Rev. 38 (2014) 717–731, doi:http://dx.doi. org/10.1016/j.rser.2014.07.040. J. Hu, S. You, M. Lind, J. Østergaard, Coordinated charging of electric vehicles for congestion prevention in the distribution grid, IEEE Trans. Smart Grid 5 (2014) 703–711, doi:http://dx.doi.org/10.1109/TSG.2013.2279007. F. Mwasilu, J.J. Justo, E.K. Kim, T.D. Do, J.W. Jung, Electric vehicles and smart grid interaction: a review on vehicle to grid and renewable energy sources integration, Renew. Sustain. Energy Rev. 34 (2014) 501–516, doi:http://dx.doi. org/10.1016/j.rser.2014.03.031. K. Knezovi c, M. Marinelli, P. Codani, Y. Perez, Distribution grid services and flexibility provision by electric vehicles: a review of options, Power Engineering Conference (UPEC), 50th International Universities, 1–4 Sept. 2015, 2015, pp. 1–6. C. Gouveia, C.L. Moreira, J.A.P. Lopes, D. Varajao, R.E. Araujo, Microgrid service restoration: the role of plugged-in electric vehicles, IEEE Industrial Electronics Magazine, vol. 7, no. 4, Dec., 2013, pp. 26–41, doi:http://dx.doi.org/10.1109/ MIE.2013.2272337. W. Kempton, J. Tomi c, Vehicle-to-grid power implementation: from stabilizing the grid to supporting large-scale renewable energy, J. Power Sources 144 (2005) 280–294, doi:http://dx.doi.org/10.1016/j.jpowsour.2004.12.022. S. Martinenas, M. Marinelli, P.B. Andersen, C. Træholt, Implementation and demonstration of grid frequency support by V2G enabled electric vehicle, Proc 49th Int Univ Power Eng Conf IEEE (2014) 1–6, doi:http://dx.doi.org/10.1109/ UPEC.2014.6934760. S. Izadkhast, P. Garcia-Gonzalez, P. Frias, An aggregate model of plug-in electric vehicles for primary frequency control, IEEE Trans. Power Syst. 30 (2015) 1475–1482, doi:http://dx.doi.org/10.1109/TPWRS.2014.2337373. H. Liu, Z. Hu, Y. Song, J. Lin, Decentralized vehicle-to-grid control for primary frequency regulation considering charging demands, IEEE Trans. Power Syst. 28 (2013) 3480–3489, doi:http://dx.doi.org/10.1109/TPWRS.2013.2252029. P.M.R. Almeida, J.A.P. Lopes, F.J. Soares, L. Seca, Electric vehicles participating in frequency control: operating islanded systems with large penetration of renewable power sources, 2011 IEEE PES Trondheim PowerTech Power Technol a Sustain Soc POWERTECH 1 (2011) 5–10, doi:http://dx.doi.org/ 10.1109/PTC.2011.6019424.
73
[18] T. Masuta, A. Yokoyama, Supplementary load frequency control by use of a number of both electric vehicles and heat pump water heaters, IEEE Trans. Smart Grid 3 (2012) 1253–1262, doi:http://dx.doi.org/10.1109/ TSG.2012.2194746. [19] J. Meng, Y. Mu, H. Jia, J. Wu, X. Yu, B. Qu, Dynamic frequency response from electric vehicles considering travelling behavior in the Great Britain power system, Appl. Energy 162 (2016) 966–979, doi:http://dx.doi.org/10.1016/j. apenergy.2015.10.159. [20] J. Druitt, W.-G. Früh, Simulation of demand management and grid balancing with electric vehicles, J. Power Sources 216 (2012) 104–116, doi:http://dx.doi. org/10.1016/j.jpowsour.2012.05.033. [21] X. Luo, S. Xia, K.W. Chan, A decentralized charging control strategy for plug-in electric vehicles to mitigate wind farm intermittency and enhance frequency regulation, J. Power Sources 248 (2014) 604–614, doi:http://dx.doi.org/ 10.1016/j.jpowsour.2013.09.116. [22] P.B. Andersen, M. Marinelli, O.J. Olesen, C.A. Andersen, G. Poilasne, B. Christensen, et al., The Nikola project intelligent electric vehicle integration, IEEE PES Innov Smart Grid Technol Eur (2014) 1–6, doi:http://dx.doi.org/ 10.1109/ISGTEurope.2014.7028765. [23] ENTSO-E, Continental Europe Operation Handbook. Available: https://www. entsoe.eu/publications/system-operations-reports/operation-handbook/ Pages/default.aspx. [24] Energinet.dk, Ancillary services to be delivered in Denmark—Tender conditions. Available: https://energinet.dk/EN/Soeg/Sider/ Publikationsvisning.aspx?Filter1Value=4812&Filter1Field=Id. [25] M. Marinelli, F. Sossan, G.T. Costanzo, H.W. Bindner, Testing of a predictive control strategy for balancing renewable sources in a microgrid, Sustain. Energy IEEE Trans. 5 (2014) 1426–1433, doi:http://dx.doi.org/10.1109/ TSTE.2013.2294194. [26] P. Kundur, Power System Stability and Control, McGraw-Hill Education, 1994. [27] L.L. Grigsby, Power System Stability and Control. The Electric Power Engineering Handbook, CRC Press/IEEE Press, 2001, 2016. [28] V. Lakshmanan, M. Marinelli, J. Hu, H.W. Bindner, Provision of secondary frequency control via demand response activation on thermostatically controlled loads: solutions and experiences from Denmark, Appl. Energy 173 (2016) 470–480, doi:http://dx.doi.org/10.1016/j.apenergy.2016.04.054.